ja1r0 / dqn-play-supermario

implement the classic reinforcement learning algorithm DQN to play supermariobrother

DQN-play-SuperMario

Use the deep reinforcement learning algorithm DQN to play the game SuperMario.
The game environment is come from https://github.com/justinmeister/Mario-Level-1.git
The agent performance is not good enough.Actually there is a gap between my expectations and the fact.This kind of learning algorithms are magical and hard to control.There must be some internal logic between the agent performance and a bunch of hyperparameters.And it is important to note that the supermariobros game is complicated than Atari games in gym,the game frame is more rich and has more changes.

This project is to be continue!

Anyway,you can just clone this repository and run "play_mario.py"

requirement

• python3.5
• opencv-python you can install it using conda conda install -c menpo opencv
• pytorch
• the supermario game environment

dqn-play-supermario's Issues

训练时间时间

请问你这个模型训练周期大概是多久。用什么显卡训练的